Comparative Analysis of Deep Learning Architectures for Hourly Solar Irradiation Forecasting

Solar radiation is an important parameter to consider when designing a solar energy system and ensuring it operates at peak efficiency. Due to the high cost of purchasing and maintaining measuring instruments, solar radiation records are not available in most regions. The goal of this work is to make a prediction model that can reliably estimate solar radiation. Recurrent Neural Networks (RNN), Gated Recurrent Units (GRU), and Long Short-Term Memory (LSM) are three popular deep Learning models, and their results in predicting solar radiation are compared in this research. The models were trained on meteorological data that was gathered over the course of a year from six distinct locations in hot areas in Tamil Nadu, India. Numerical analysis shows that the GRU regression methodology is superior to the other two in predicting solar radiation. The GRU has an MAPE of 1.39% and RMSE of 0.55% during the training phase, and 2.45% and 0.89% during the testing phase, respectively. The proposed GRU model is more accurate and can be used in more similar situations. It can be used to estimate the hourly average global radiation for developing PV installations. This information is important for determining the size and efficiency of a PV installation and ensuring that a PV installation is designed to meet the specific needs of a particular location.